Research map / Label-efficient and robust learning / Test-time adaptation
Learning to Receive Help: Intervention-Aware Concept Embedding Models
Mateo Espinosa Zarlenga, Katherine M. Collins, Krishnamurthy Dj Dvijotham, Adrian Weller, Zohreh Shams, Mateja Jamnik
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| IntCEM loss (lambda_roll=1) vs standard CEM training | CEM · MNIST-Add | task AUC: 90.13 ± 0.41 -> 92.07 ± 0.20 | Improved |
| IntCEM (lambda_roll=0.1) vs CEM | CEM · MNIST-Add-Incomp | task AUC: 86.61 ± 0.75 -> 89.63 ± 0.13 | Improved |
| IntCEM (lambda_roll=1) vs CEM | CEM · CUB | task accuracy: 79.00 ± 0.61 -> 78.03 ± 0.58 | Hurt |
| IntCEM (lambda_roll=1) vs CEM | CEM · CUB-Incomp | task accuracy: 75.52 ± 0.37 -> 74.70 ± 0.22 | No clear effect |
| IntCEM (lambda_roll=5) vs CEM | CEM · CelebA | task accuracy: 31.04 ± 0.96 -> 38.09 ± 0.26 | Improved |
| Lower lambda_roll (5 -> 0.1) in IntCEM (ablation) | IntCEM · CUB | mean concept AUC: 89.63 ± 0.42 -> 93.39 ± 0.22 | Improved |
| IntCEM (lambda_roll=5) vs CEM | CEM · CelebA | mean concept AUC: 85.28 ± 0.18 -> 88.25 ± 0.24 | Improved |
| Joint CBM-Logit vs CEM on concept prediction | CEM · MNIST-Add | mean concept AUC: 85.51 ± 0.26 -> 89.21 ± 0.71 | Hurt |
| Random test-time interventions on IntCEM vs baselines | IntCEM · CUB | task accuracy: ~10% with ~25% of concepts intervened | Improved |
| Test-time policy: learned psi vs CooP query time | IntCEM · CUB | policy query time (ms): 4.77 ± 0.10 -> 2.90 ± 0.20 | Improved |
| Learn psi end-to-end instead of post-hoc behavioural cloning (BC-Skyline) (ablation) | IntCEM · CUB | task accuracy vs interventions | Improved |
| Remove learned train-time policy psi (random trajectories, IntCEM no psi) (ablation) | IntCEM · CUB | task accuracy vs interventions | Hurt |
Closest papers by meaning
- Learning to Intervene on Concept Bottlenecks (ICML 2024 · similarity 0.92)
- Stochastic Concept Bottleneck Models (NeurIPS 2024 · similarity 0.89)
- Causally Reliable Concept Bottleneck Models (NeurIPS 2025 · similarity 0.89)
- Sample-efficient Learning of Concepts with Theoretical Guarantees: from Data to Concepts without Interventions (NeurIPS 2025 · similarity 0.88)
- A theoretical design of concept sets: improving the predictability of concept bottleneck models (NeurIPS 2024 · similarity 0.88)
- Energy-Based Concept Bottleneck Models: Unifying Prediction, Concept Intervention, and Probabilistic Interpretations (ICLR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
